Unverified50% confidenceFactExact time
PR-AUC 相比 ROC-AUC 不会被大量负样本的正确分类所稀释，更适合稀有事件检测场景
1
Sources
50%
Confidence
Long-term
Relevance
8/14/2026
First Seen
Sources
Related Claims
UnverifiedSecurity issues in large language models currently lack a standardized classification system analogous to CVE, and there is no unified standard for assessing vulnerability severity.57% similarUnverifiedRepresentative works in safety evaluation datasets include AdvBench, HarmBench, and SALAD-Bench, and a unified standard like MMLU has yet to be established57% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/746398API
curl https://kongchang.com/api/v1/knowledge/claims/746398MCP
get_claim(id=746398)